Jul 10, 2023 06:00:00

Specific language specialized language model 'InternLM' that can demonstrate higher performance than GPT-4

While the high-performance chat AI 'ChatGPT' is attracting a lot of attention, language models that appeal to performance comparable to ChatGPT are being announced one after another. The language model ' InternLM ' developed mainly by the Shanghai AI Research Institute has been trained specifically for English and Chinese, and has recorded scores exceeding ChatGPT in multiple performance evaluation tests.

Intern LM

https://internlm.org/

InternLM is a language model jointly developed by the Shanghai AI Research Institute and multiple universities and companies, trained with over 1 trillion tokens of multilingual data. The main features of InternLM are as follows.

・Have the ability to understand and express multiple languages, and demonstrate outstanding performance in English and Chinese in particular.
・Excellent performance in fields such as logical reasoning and code generation.
- Exceeded ChatGPT in various Chinese performance evaluation tests and demonstrated performance close to GPT-4.

In Microsoft's language model performance evaluation test ' AGIEval ', InternLM (blue) outperforms ChatGPT (red) in all elements. In addition, 'GK-En', 'GK-history' and 'GK-Cn' have scored higher than GPT-4 (light blue).

The results of '

C-Eval ' that can evaluate the Chinese processing performance of the language model are as follows. InternLM (blue) outperforms GPT-4 (light blue) on most tests except 'Middle School Biology'.

InternLM can communicate in a chat format like ChatGPT, and can also generate code according to written instructions.

InternLM model data is published in the following GitHub repository.

GitHub - InternLM/InternLM: InternLM has open-sourced a 7 billion parameter base model, a chat model tailored for practical scenarios and the training system.
https://github.com/InternLM/InternLM